IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 46, NO.

5, MAY 2001 711

A Parametric Solution to the Pole Assignment

Problem Using Dynamic Output-Feedback
Mehmet Turan Söylemez, Associate Member, IEEE, and Neil Munro

Abstract—A new technique is presented for pole placement of 1) What is the minimum order of the compensator , for
linear time-invariant systems using dynamic feedback. A recently which a solution to the pole assignment problem exists?
developed method for partial pole assignment using constant feed-
2) What is the maximum number of poles that can be as-
back is generalized to the dynamic output-feedback case. Subject
to a mild assumption on the number of complex conjugate poles signed with a compensator of order ?
to be assigned, it is almost always possible to arbitrarily assign all Since Wonham’s seminal work in 1967 [2], where the
the closed-loop system poles using a compensator of order ( relationship between controllability and pole assignment was
) max( ) using this new method. Here, , , and are established, there has been a lot of work on the pole assignment
the order of the system, and the number of inputs and outputs, re-
spectively, and problem. Pearson and others were among the first who studied
the output-feedback pole assignment using dynamic compen-
max( ) max( ) sators [3]–[5] (1969/70). They showed the relationship between
max( )+ + +
2 min( ) the minimum order of the pole assignment compensator and
where denotes the nearest integer lower than or equal to the controllability and observability indices of the system.
(i.e., floor ( )), and denotes the nearest integer greater than Chen [6] (1970), and then Chen and Hsu [7] (1971) gave a
or equal to (i.e., ceiling ( )). An equivalent result is that using a method for finding dyadic dynamic compensators for cyclic
compensator of order , it is almost always possible to arbitrarily systems. This method was improved in Seraji [8] (1975) to
assign min( + (max( ) + 1) + ) closed-loop system
poles. Only the normal procedures of linear algebra are required to cover both complete and partial pole assignment, and in Munro
implement the technique. Note that + 1 and, therefore, and Novin–Hirbod [9] to noncyclic systems and full-rank
the result is stronger than previous exact pole assignment results. compensators. Motivated by Pearson’s [4], [5] and Davison’s
Since it does not involve iteration or any other numerical tech- [10] work, Ahmari and Vacroux [11] (1973) gave a sufficient
niques, it is possible to implement the method symbolically and,
therefore, to obtain general parametric solutions to the pole assign-
condition and formula for the number of poles to be assigned by
ment problem. The freedom in this design approach can also often a fixed order compensator. Patel, on the other hand, extended
be used to guarantee the internal stability and/or robustness of the the work done by Fallside and Seraji [12] for constant output
resulting closed-loop system. feedback pole assignment to the dynamic output-feedback case
Index Terms—Dynamic output feedback, pole assignment, pole in 1976 [13]. Kimura [14] (1975), in his paper on static-output
retention. feedback pole assignment, stated that a compensator of order ,
such that , would be adequate for complete
I. INTRODUCTION pole assignment. Munro and Novin–Hirbod [9] (1979) reduced
this number by a factor of . Seraji later [15] (1980)

T HE dynamic output feedback pole assignment problem

can be stated as follows. Given a system
confirmed this result by using a two-step algorithm.
As the methods using linear algebra hit the barrier provided
by Munro and Novin–Hirbod, a new area of research called the
geometric algebraic approach became more popular day by day,
(1) especially in the search for the necessary and sufficient condi-
where , and and a set of tions for generic1 pole assignment by output-feedback. Using
arbitrarily chosen complex numbers , geometric algebra arguments, Hermann and Martin [16] proved
closed under complex conjugation, find the output-feedback in 1977 that if generic pole assignability holds for the
compensator, , of order such that the set of closed-loop class of static feedback controllers (i.e., ), provided com-
system eigenvalues are equal to . Two related challenges plex feedback matrices are allowed in the closed-loop system.
arising from this problem have attracted much attention in the In the following year, Willems and Hesselink [17] showed that
pole assignment literature. is a necessary condition for the
pole assignment of generic systems using (real) output-feed-
back. These seminal results sparked an extensive research on
Manuscript received April 15, 1998; revised August 27, 1999, July 11, 2000,
and November 30, 2000. Recommended by Associate Editor S. Weiland. 1In the generic pole assignment problem, the definition of the (exact) pole
M. T. Söylemez is with the Department of Electrical Engineering, Istanbul assignment problem is “softened” in the sense that instead of starting from a
Technical University, Istanbul, Turkey (e-mail: [email protected]). given plant the set of all possible plants with given n, m, and `, minus possibly
N. Munro is with the Control Systems Centre, UMIST, Manchester, U.K. a variety in this set is considered, and minimum order of compensators that
(e-mail: [email protected]). assign the closed-loop system poles arbitrarily close to any given set of desired
Publisher Item Identifier S 0018-9286(01)03621-2. pole locations (0) is being sought.

0018–9286/01$10.00 ©2001 IEEE

Authorized licensed use limited to: WASEDA UNIVERSITY LIBRARY. Downloaded on August 28,2023 at 05:26:03 UTC from IEEE Xplore. Restrictions apply.
712 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 46, NO. 5, MAY 2001

the generic pole assignment problem. Most of these earlier re- back pole assignment using constant compensators, showing
sults as summarized by Byrnes [18], however, assumed the im- that it is possible to assign poles of the closed-loop
plementation of complex compensators in the feedback path, system using a constant compensator. In this paper, this result is
and therefore, were impractical. The real breakthrough in this generalized to the dynamic output-feedback case and it is shown
area of research was in 1992, when Wang [19] showed that that, subject to a mild assumption (see Assumption IV.1 and
is a sufficient condition for generic pole assignment Remark 2), it is possible to assign
using (real) static output-feedback.2 Dynamic feedback coun- poles using a compensator of order or, equiva-
terparts of this result (under the assumption of allowing com- lently, to assign all the closed-loop system poles using a com-
plex feedback compensators) followed [21], [22]. pensator of order . This provides an
As Kimura [23] stated in 1994, one of the main problems improvement over the previous exact pole assignment results,
with the geometric algebra methods was the unavailability of when and .
practical algorithms to find the required compensators. More The method to be presented is a multi-step method, where
recently several algorithms have been suggested to find the in each step some poles of the closed-loop system are assigned
required compensators for generic pole assignment. One such using an input (or an output) of the system and are then made un-
algorithm was due to Leventides and Karcanias [24] for the controllable (unobservable) from other inputs (outputs) so that
constant output-feedback case. This algorithm, however, used the following steps do not affect the previously assigned poles.
asymptotically infinite gains in the feedback path, and therefore In order to lay the foundations of the method, a partial pole
was not very practical from the engineering point of view. Wang assignment technique for single-input systems is introduced in
[25] (for the constant feedback case) and Rosenthal and Wang Section II. Here, it will be shown that it is possible to param-
[26] (for the dynamic feedback case) provided iterative algo- eterize the solution such that the extra degrees of freedom in
rithms to find the required pole assignment compensators. The a design are kept as free parameters. The problem of how to
algorithm provided by Rosenthal and Wang [26], in particular, use these free parameters to obtain pole retention is discussed
would allow generic pole assignment for a system provided that in Section III. The main result is summarized, and an easy to
. Here, is a generally-small implement construction technique for the required pole assign-
integer defined as , where ment technique is provided in Section IV. Section V presents an
and are the remainders of divided by and , respectively. example, where the application of the method to a robust control
Considering that when is a multiple of or it provides the problem is illustrated. Finally, concluding remarks are given in
best possible bound for generic problems, this result is very Section VI.
impressive. There are, however, several disadvantages of the
algorithm provided. To start with, the algorithm is based on a II. PARTIAL POLE ASSIGNMENT FOR SINGLE-INPUT SYSTEMS
Newton–Raphson type iteration the success of which depends USING AN OUTPUT FEEDBACK COMPENSATOR
on a small scalar parameter ( ). As the authors point out, if
Consider the single-input system
is too small, some coefficients of the compensator can be very
high. Although a second iteration on is suggested to prevent
high gains in the feedback, where to stop this second iteration
(2)
is not certain. Another, and probably more important, problem
with the implementation of the resulting compensators, is that where , and . Now, suppose
the gains determined have to be implemented with very high it is required to find a compensator, , of order such that
accuracy. (Actually, Yang and Tits [27] showed in 1998 that of the closed-loop system poles are to be equal to a complex
generic pole assignment may produce very fragile designs.) conjugate set of numbers, , where
Furthermore, the use of iteration prevents the parameterization
of solutions, when extra degrees of design freedom exist. (3)
However, such a parameterization can be very desirable if
there are further design specifications to be considered such In this section, it will be shown that the required compensator
as eigenvector assignment, robustness (see [28]), or internal can be obtained such that extra degrees of
stability. As a result, although geometric algebra has proved to freedom in the design are kept. In order to do this, the relation-
be extremely useful in determining the necessary and sufficient ship between the coefficients of the compensator and those of
conditions for the generic pole assignment problem, the com- the closed-loop system characteristic polynomial is first found.
pensators found using the resulting algorithms are usually very Note that the transfer function matrix of the given system can
difficult, if not impossible, to implement. be calculated from
As far as the linear algebra methods concerned, on the other
hand, the limit provided by Munro and Novin–Hirbod [9] (i.e., (4)
) has remained as the limit for
minimal order (exact) pole assignment. Only recently, Söylemez
and Munro [1] have introduced a method for partial output-feed- (5)
..
.
2A simpler proof of this result is given by Willems [20].
Authorized licensed use limited to: WASEDA UNIVERSITY LIBRARY. Downloaded on August 28,2023 at 05:26:03 UTC from IEEE Xplore. Restrictions apply.
SÖYLEMEZ AND MUNRO: A PARAMETRIC SOLUTION TO THE POLE ASSIGNMENT PROBLEM 713

where is the identity matrix, is the open-loop where is the th. column of and is the th. row of .
system characteristic polynomial, defined as Then

(6)
(7)
.. .. .. ..
. . . .
and is an matrix formed by the coefficients of the numer-
ator part of the transfer function matrix. When the given system
is controllable, an alternative method of finding is by using
the following formula: .. .. ..
. . .

(8) (15)

where is the similarity transformation matrix that transforms Now, writing the equations in block matrix form and equating
the given system into the controllable standard form, and can be coefficients of like powers of it is possible to show that
calculated using any of the methods mentioned in [29].
Now, assuming that the compensator has the form
(16)

(9) where, is the difference vector formed by the coefficients of

the difference polynomial , i.e.,

where is the characteristic polynomial of the compensator

(17)
(10) is a matrix formed as (18) shown at
the bottom of the next page, and is a vector that contains the
parameters of the compensator given as
and , the closed-loop system characteristic poly-
nomial is then given by
(19)
(11)
(12) where

(20)

.. and
.
(21)
(13)

.. Therefore, the required relationship between the coefficients

. of the compensator and the closed-loop system characteristic
polynomial is given by (16). We note that a similar formulation
to this equation is given in [7] and [9]. We also note that (16) is
consistent for the constant output feedback case ( ), where
becomes the difference vector formed by the coefficients of
.. , reduces to the transpose of and becomes a
.
constant output feedback vector.
Equation (16) must now be rearranged such that all the
unknowns in this equation (i.e., the unknown part of the
closed-loop system characteristic polynomial and some or all
.. coefficients of the compensator) are expressed by what is given
. in the problem and some free parameters. In order to do this, the
closed-loop system characteristic polynomial is first separated
(14) into two parts by defining two polynomials , and ,
Authorized licensed use limited to: WASEDA UNIVERSITY LIBRARY. Downloaded on August 28,2023 at 05:26:03 UTC from IEEE Xplore. Restrictions apply.
714 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 46, NO. 5, MAY 2001

where is the achievable part of the desired closed-loop where

characteristic polynomial given by formed by the coefficients of ;
vector formed by the unknown coefficients ;
is
(22)

(32)
(23)
where the are formed by the coefficients of .
Using (31) in (16), we obtain
and is the “residue polynomial” formed by the rest of the
closed-loop system poles, i.e., (33)

(24) Now, so as to extract the extra degrees of freedom in the design,

is partitioned into two parts
where
(34)
(25)

Then, we can write where contains elements and contains

free elements. Partitioning the matrix accordingly as
(26)
(35)
(27)
we can write
(28)
(36)

The difference polynomial is then

Hence
(29)
(37)
(30)

(38)
Hence, it is possible to write

(31) (39)

.. .. .. .. .. .. .. .. .. .. .. ..
. . . . . . . . . . . .
.. .. .. .. .. ..
. . . . . .
.. .. .. .. ..
. . . . .
.. .. .. (18)
. . .
.. .. .. .. ..
. . . . .
.. .. .. .. .. .. .. .. ..
. . . . . . . . .
.. .. .. .. .. ..
. . . . . .

Authorized licensed use limited to: WASEDA UNIVERSITY LIBRARY. Downloaded on August 28,2023 at 05:26:03 UTC from IEEE Xplore. Restrictions apply.
SÖYLEMEZ AND MUNRO: A PARAMETRIC SOLUTION TO THE POLE ASSIGNMENT PROBLEM 715

Noting that is a square matrix, and is generally3 nonsingular From (34)–(40), the pole assignment compensator that as-
if is full rank and has at least controllable and signs these poles of the closed-loop system can be given as
observable modes, the solution to (39) is given by

(42)
(40)

(43)
Hence, elements in the second part of the compensator pa-
rameter vector ( ) and coefficients of the
residue polynomial can be found in terms of what has been (44)
given in the problem and some free variables in . We note
that if , then includes all the coefficients of
the characteristic polynomial of the compensator. Therefore, in where is an identity matrix and
parallel with pole assignment, it is possible to determine the
internal dynamics of the system arbitrarily and hence guarantee
the internal stability of the system in such a situation. In the
following section, we shall show that it is possible to use the (45)
freedom provided by for pole retention in the constant output
feedback case.
Hence, if we partition the matrix appropriately as
III. POLE RETENTION
As was mentioned earlier, the results provided in Section II (46)
are valid when (i.e., the constant feedback case). In this
section, we shall consider a controllable and observable multi- we can write
input–multi-output (MIMO) system as given in (1) with ,
full rank and for nontriviality, assume4 that ,
and show that it is possible to assign and retain some poles using (47)
a constant output feedback compensator applied only to the first
input of this system. This result is summarized in the following (48)
theorem, which is taken from [1].
Theorem III.1: By using a constant feedback compensator
from the system’s outputs to the first input of the system given where
in (1), it is almost always possible to arbitrarily assign up to
self conjugate poles of the system, and make
these poles uncontrollable from the other inputs, provided that
(49)
the system has controllable and observable poles,
where is the first column of the input matrix ( ).
Proof: Proving the first part of the theorem (pole assign- We also know that, in order to make a pole (say ) uncon-
ment) is obvious due to the results provided in the previous trollable from the rest of the system’s inputs, has to
section. We shall denote the constant feedback compensator satisfy
as ( (for )), and define and
for clarity in the subsequent notation. In the following, we
shall show that the free variables that occur in the pole assign- Adj for (50)
ment step ( ) can be used to make the assigned closed-loop
system poles uncontrollable from the other inputs (pole reten-
tion). where is the identity matrix, is the th. column of
We know that the resulting closed-loop system matrix is the and Adj denotes the adjoint matrix. Moreover, we know
given by that the rank of the adjoint matrix in (50) is 1[15].Therefore, (50)
can be satisfied by loosing degrees of freedom if
(41) enters linearly into the equations. The following lemma states
this, and the proof provides a method for determining the free
parameters to guarantee uncontrollability from the inputs
3Iffor such a given system X ^ is singular, then by switching the columns of other than the first one.
^ matrix as stated in [8].
Z it is always possible to obtain a nonsingular X Lemma III.1: A solution for the free parameters in , using
4Note that this assumption does not effect the generality of the result, since if
(50) in order to make the mode uncontrollable from all the
m > ` then everything said in the following can be applied to the dual system
[A ; C ; B ]and the compensator found in the end can be transposed to find
inputs except the first input, implies solving a set of linear equa-
the pole assignment compensator for the original system. tions.
Authorized licensed use limited to: WASEDA UNIVERSITY LIBRARY. Downloaded on August 28,2023 at 05:26:03 UTC from IEEE Xplore. Restrictions apply.
716 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 46, NO. 5, MAY 2001

Proof: Let us denote the element of Adj on any nonzero row of the matrix (denoting this as ), and
the th. row and th. column as , then require that satisfies

Adj (51) (61)

Adj (52)
(53) A vector that satisfies (61) ensures that the mode is not
controllable from th. input in the resulting closed-loop system.
where Then, in order to make all poles uncontrollable from all the
matrix without the th. column and -th other inputs has to satisfy
row;
vector without the -th element;
matrix without the -th column.
However (62)
Note that this implies linear equations have to be
satisfied by . However, the length of is greater than or
Adj (54) equal to the number of equations; i.e.,

Now, it is obvious that enters into the elements of the adjoint (63)
matrix linearly, hence, the proof is complete.
In fact, if we denote as Therefore, we can almost always find a solution for to sat-
isfy (62) and the proof of the theorem is complete. Note, how-
Adj (55) ever, that if poles with multiplicity greater than one are to be
assigned, although it is possible to assign these it is not possible
to retain them with the current method. So, multiple poles need
then the condition in (50) implies to first be perturbed slightly if they have to be retained. This cor-
responds to the “almost” part in the theorem.
We remark that if there are some complex-conjugate eigen-
.. values to be assigned the calculations have to be done using
Adj . (56)
complex arithmetic. This is not a problem if a computation envi-
ronment that supports operations on complex matrices such as
Matlab or Mathematica is being used. Otherwise, care should
for , where be taken. If for example and are two complex conjugate
poles to be assigned, then the vectors and ( )
turn out to be complex conjugate and, therefore, it is possible to
(57) use Re and Im instead of these vectors in (62).

Hence, has to satisfy the following equations: IV. POLE PLACEMENT FOR MIMO SYSTEMS
Let us define variables, , , , , where

Adj .. (58)
. (64)

where denotes the nearest integer lower than or equal to .

.. (59) Following the arguments in the preceding sections, it is now pos-
.
sible to establish the following theorem, which forms the main
contribution of this paper. Before giving this theorem, however,
let us give the assumption that was mentioned in the introduc-
tion section first.
(60)
Assumption IV.1: We assume that the set of desired
closed-loop poles, , can be partitioned into
disjoint subsets such that
Now, since the rank of the adjoint matrix in (50) is one, the
rank of the matrix is also one. Therefore, we can pick up (65)
Authorized licensed use limited to: WASEDA UNIVERSITY LIBRARY. Downloaded on August 28,2023 at 05:26:03 UTC from IEEE Xplore. Restrictions apply.
SÖYLEMEZ AND MUNRO: A PARAMETRIC SOLUTION TO THE POLE ASSIGNMENT PROBLEM 717

where each subset ( ) has elements closed under complex In order to see that a compensator of order is enough to
conjugation, and satisfies the following: assign poles, we consider the fact that (from (66))

(66) (68)

for (67)
and, since due to (67), it is possible to write that
Moreover, it is assumed that multiple poles exist only in .
Theorem IV.1: (Main Result) Given a completely control-
(69)
lable and observable system as in (1) with and matrices
full rank, arbitrary pole assignment is always possible using a
compensator of order , under the as-
sumption that the desired closed-loop system poles can be par- It is also known that is by definition . Hence
titioned as defined in Assumption IV.1.
Proof: In the following, it will be assumed that . (70)
This does not affect the generality of the result, since, as men-
tioned earlier, when the dual system can be On the other hand, it is known that using a compensator of
considered to satisfy this assumption. order it is possible to assign at most poles for
The proof is given by construction: In order to construct the single input systems. Hence
desired compensator we use the following -step algorithm.
Step 1) Here, if the system has at least controllable
modes through the first input, following directly (71)
from Theorem III.1 a constant compensator applied
to this input can be used to assign poles in Therefore has to satisfy the following inequality:
and make these poles uncontrollable from the other
inputs, since . We can then find the feed- (72)
back law to be applied to the first input, or in other
words, the first row of the transfer function ma-
trix of the output feedback compensator using (44), i.e.,
where is determined from (62). If it is not pos-
sible to control modes through the first input (73)
then another input can be selected instead. If it is
(74)
not possible to assign this many poles from any of
the inputs, then it is always possible to make the
system controllable from any of its inputs by using The smallest integer that satisfies (74) is ,
an almost arbitrary compensator in the feedback hence, the proof is complete.
before starting the pole assignment procedure [10]. Remark 1: It is also possible to prove that by using the
Step 2) We now consider the closed-loop system obtained technique presented above
in the previous step without the first input. There- closed-loop poles can almost always be arbitrarily assigned
fore, an -th order system with inputs, when a compensator of order is used in the feedback. Note
outputs and uncontrollable modes is considered. that this is consistent with the result provided in [1] for partial
Following a similar procedure to that given in step pole assignment using constant output feedback ( ).
one, it is now possible to assign poles in and Remark 2: The assumption made in the above theorem is a
make these poles uncontrollable from the rest of the “mild” assumption in the sense that in many cases it has no ef-
inputs using the second input of the original system. fect other than stating that the set of desired closed-loop system
The poles assigned in the first step are not effected poles should be closed under complex conjugation. In some
from this assignment, and they remain uncontrol- cases, however, it may not be possible to select all closed-loop
lable from the other inputs, since they are uncon- system poles as complex conjugate pairs, and some poles may
trollable through the second input. have to be selected from the real field.
..
. Remark 3: Making several subsystems uncontrollable may
Step m) The same procedure can be used up to step m. Here, seem to be unusual at first sight, and one may think that this
since there are no further inputs to be considered, might not be very practical. Although it is true that in a prac-
there is no need for pole retention, and following tical system it may not always be possible to obtain exact un-
the discussions in Section II we can use a dynamic controllability, a near uncontrollability condition can always be
single input compensator of order achieved. This would be sufficient to make sure that the poles
to assign the remaining poles in . assigned in the earlier steps are not affected much by the later
Authorized licensed use limited to: WASEDA UNIVERSITY LIBRARY. Downloaded on August 28,2023 at 05:26:03 UTC from IEEE Xplore. Restrictions apply.
718 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 46, NO. 5, MAY 2001

steps of the design process. One should also note that the idea If , in particular, all the poles of the
of pole retention is not new, and all the earlier multistep linear compensator can be assigned arbitrarily.
algebra methods (including those of [9] and [15]) make a sub- In the following section, an example is given to illustrate the
system uncontrollable (or unobservable) one way or another. new method in a robust pole assignment framework.
Remark 4: The method and specially the pole retention algo-
rithm provided by (51)–(62) might seem to be computationally V. EXAMPLE
intensive. However, it is not difficult to show that the compu- Consider the open-loop unstable system, shown in (75) and
tational complexity of the algorithm is of polynomial type. We (76) at the bottom of the page, where the ( ) are
particularly note that it is generally sufficient to calculate just uncertain parameters with nominal values zero and are bounded
one row of the matrix defined in (60). This reduces the com- as .
putational requirements of the algorithm dramatically. In fact, if The poles of the open-loop nominal system are
a symbolic algebra language is being used the direct solution of
(50) can be even faster. For most of the practical problems we
considered, these calculations take less than a second.
Remark 5: We note that when the order of the system is large, (77)
the method can introduce some numerical errors due to the fact
that the matrix in (40) is generally a poorly conditioned ma- and it is required to design a first-order dynamic-feedback com-
trix. This, however, can be overcome if a symbolic algebra lan- pensator with a pole at to assign the nominal closed-loop
guage is used. In fact, using such direct methods usually gives system poles to
better results in symbolic computation environments, as noted
in [30]. (78)
Remark 6: Since the method introduced does not involve any
numerical techniques such as iteration, with the help of a sym- Let us assume that it is also required that the pair of domi-
bolic algebra language (see [28] and [31]), it is also possible nant poles of the closed-loop system lie in disks centered at
to find parametric solutions to the pole assignment problem with radius 0.8 under all possible perturbations. Further-
when extra degrees of freedom are available (i.e., when more, it is required that for all the other complex poles of the
). This is especially important when further de- closed-loop system the distances between the perturbed poles
sign requirements such as robustness (see [32] and [33]), eigen- and the nominal poles are less than 1.2, and that the real pole of
vector assignment, and internal stability are to be satisfied. In the closed-loop system is to be bounded by .
order to emphasize this last point; namely, that internal stability Söylemez and Munro [28] have shown that such a robust pole
requirements can be satisfied with the current method; we re- assignment problem can be considered in two parts. First, the
mark that the free parameters provided by the vector in (34) set of pole assignment compensators are found for the nominal
contain some or all of the coefficients of the characteristic poly- plant, and then a robust solution is selected from this set. Note
nomial of the compensator. Therefore, by selecting these free that the iterative method of Rosenthal and Wang [26] does not
parameters suitably it is possible to guarantee internal stability. allow this parameterization. Also, the earlier linear algebra

(75)

(76)

Authorized licensed use limited to: WASEDA UNIVERSITY LIBRARY. Downloaded on August 28,2023 at 05:26:03 UTC from IEEE Xplore. Restrictions apply.
SÖYLEMEZ AND MUNRO: A PARAMETRIC SOLUTION TO THE POLE ASSIGNMENT PROBLEM 719

methods of Munro and Novin-Hirbod [9] and Seraji [15] do and from (31) and (32)
not provide enough design freedom to both assign the pole of
the compensator and obtain a parametric solution. Therefore,
the method just introduced provides a new way of determining
a parametric solution to the given nominal pole assignment
problem. To show this, we first consider the nominal system. If
is now partitioned as where (82)

(79)
(80)

then the poles in can be assigned in Step 1) and the rest of

the poles can be assigned in Step 2).
Hence, see (83), shown at the bottom of the page, and
Step 1: From (18)

(84)
(81)

and are then calculated from (49) as (85) and (86), shown
at the bottom of the page. Further calculations yield the matrix

(83)

(85)

(86)

Authorized licensed use limited to: WASEDA UNIVERSITY LIBRARY. Downloaded on August 28,2023 at 05:26:03 UTC from IEEE Xplore. Restrictions apply.
720 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 46, NO. 5, MAY 2001

in (62) as (87), shown at the bottom of the page. Then, a Noting that the first of these parameters directly determines the
that satisfies (62) is calculated as pole of the compensator, we select

(88)
(93)

and, therefore, the first row of the feedback matrix is determined

as Then, following the procedure used in Step 1, it is possible to
show that (94), shown at the bottom of the page, holds true and
(89)
(90)

Step 2: In this step, we assign rest of the desired closed-

loop system poles. For this, we now consider the system (95)
, where (91), shown at the bottom of the page,
holds true, and the is the th column of the matrix. We
note that since only seven poles in are to be assigned at this
step the vector defined by (34) contains two free parameters.

(87)

(91)

(92)

(94)

Authorized licensed use limited to: WASEDA UNIVERSITY LIBRARY. Downloaded on August 28,2023 at 05:26:03 UTC from IEEE Xplore. Restrictions apply.
SÖYLEMEZ AND MUNRO: A PARAMETRIC SOLUTION TO THE POLE ASSIGNMENT PROBLEM 721

Fig. 1. Pole spread of the closed-loop system for the example.

hence at the bottom of the page.The necessary overall compensator is

then as in (98), shown at the bottom of the page. We remark here
that can be selected freely to satisfy specifications on the nom-
inal pole assignment part of the problem. In order to satisfy the
specifications on robustness, the general form of the closed-loop
(96) system characteristic polynomial is found and an optimization
over is done as explained in [33]. After such an optimization,
it is found that for a robustly stabilizing controller is
obtained (see Fig. 1). Hence, the following compensator shown
in (100) at the bottom of the page, with dynamic degree deter-
The feedback law from the system outputs to the second input of mined by this approach as one, is suggested as a solution to the
the system is therefore given in parametric form by (97), shown given problem.

(97)

(98)

(99)

(100)

Authorized licensed use limited to: WASEDA UNIVERSITY LIBRARY. Downloaded on August 28,2023 at 05:26:03 UTC from IEEE Xplore. Restrictions apply.
722 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 46, NO. 5, MAY 2001

VI. CONCLUSION [14] H. Kimura, “Pole assignment by gain output feedback,” IEEE Trans.
Automat. Contr., vol. AC-20, pp. 509–516, 1975.
It has been shown that using a th. order [15] H. Seraji, “Design of pole-placement compensators for multivariable
compensator, it is almost always possible to assign all of the systems,” Automatica, vol. 16, pp. 335–338, 1980.
[16] R. Hermann and C. F. Martin, “Applications of algebraic geometry to
closed-loop system poles arbitrarily. A new technique has been systems theory—part I,” IEEE Trans. Automat. Contr., vol. AC-22, pp.
presented for the development of such a compensator. A dual re- 19–25, Jan 1977.
sult is that it is also possible to assign [17] C. Willems and H. Hesselink, “Generic properties of the pole place-
ment problem,” in Proc. IFAC Triennial World Congress, 1978, pp.
closed-loop system poles using a th-order compen- 1725–1729.
sator in the feedback path. The resulting compensator is in a [18] C. I. Byrnes, “Pole assignment by output feedback,” in Three Decades
special form having dynamic elements in only one of the rows of Mathematical System Theory, H. Nijmeijer and J. M. Schumacher,
Eds., 1989, pp. 31–78.
(or columns). This can be an advantage of the method presented [19] X. Wang, “Pole placement by static output feedback,” J. Math. Syst.
if one particular input (or output) of the system is known to be Estimat. Contr., vol. 2, no. 2, pp. 205–218, 1992.
less sensitive compared with others. In such a case, the input (or [20] J. C. Willems, “Generic eigenvalue assignability by real memoryless
output feedback made simple,” in Communications, Computation, Con-
output) with least sensitivity is considered in the last step of the trol and Signal Processing, A. Paulraj, V. Roychowdhury, and C. D.
MIMO pole assignment method. Finding a suitable ordering of Schaper, Eds. Norwell, MA: Kluwer, 1997, pp. 343–354.
the remaining inputs (or outputs), however, remains a topic for [21] J. Rosenthal, “On dynamic feedback compensation and compactification
of systems,” SIAM J. Control Optim., vol. 32, no. 1, pp. 279–296, 1994.
further research. [22] M. S. Ravi, J. Rosenthal, and X. Wang, “Dynamic pole assignment and
Another advantage of the method is that it is possible to guar- Schubert calculus,” SIAM J. Control Optim., vol. 34, no. 3, pp. 813–832,
antee internal stability using this technique in many cases. The 1996.
[23] H. Kimura, “Pole assignment by output feedback: A longstanding open
extra degrees of freedom available in the design can also be used problem,” in Conf. Decision Control, Lake Buena Vista, FL, 12, 1994,
to satisfy further design specifications such as eigenvector as- pp. 2101–2105.
signment and robustness. These are, however, topics for further [24] J. Leventides and N. Karcanias, “Global asymptotic Iinearization of the
pole placement map: A closed solution for the constant output feedback
research. problem,” Automatica, vol. 31, no. 7, pp. 1303–1309, 1995.
[25] X. A. Wang, “Grassmannian, central projection, and output feedback
pole assignment of linear systems,” IEEE Trans. Automat. Contr., vol.
ACKNOWLEDGMENT AC-41, pp. 786–794, June 1996.
[26] J. Rosenthal and X. A. Wang, “Output feedback pole placement with
The authors would like to thank all of the reviewers for their dynamic compensators,” IEEE Trans. Automat. Contr., vol. 41, pp.
830–843, June 1996.
useful comments and suggestions toward the improvement of [27] Y. Yang and A. L. Tits, “Generic pole assignment may produce very
the quality of this paper. fragile designs,” in Conf. Decision Control, Tampa, FL, 12, 1998, pp.
1745–1746.
[28] M. T. Söylemez and N. Munro, “Pole assignment for uncertain systems,”
in Symbolic Methods in Control System Analysis and Design, N. Munro,
REFERENCES Ed. London, U.K.: IEE, 1999, pp. 251–272.
[1] M. T. Söylemez and N. Munro, “A new technique for partial pole [29] M. T. Söylemez, “State feedback dyadic pole assignment methods: A
placement using constant output-feedback,” in Conf. Decision Control, unification Control Systems Centre Report,” UMIST, Manchester, U.K.,
Tampa, FL, 12, 1998, pp. 1722–1727. 872, 1998.
[2] W. M. Wonham, “On pole assignment in multi-input controllable linear [30] M. T. Söylemez and N. Munro, “Pole assignment and symbolic algebra:
systems,” IEEE Trans. Automat. Contr., vol. AC-12, pp. 660–665, June A new way of thinking,” in Proc. Control, vol. 9, Swansea, UK, 1998,
1967. pp. 1306–1310.
[3] B. Pearson and C. Y. Ding, “Compensator design for multivariable linear [31] N. Munro and M. T. Söylemez, “The use of symbolic algebra for uncer-
systems,” IEEE Trans. Automat. Contr., vol. AC-14, pp. 130–134, Feb. tain systems,” in Proc. Control, vol. 9, Exeter, UK, 1996, pp. 1332–1337.
1969. [32] M. T. Söylemez and N. Munro, “A note on pole assignment in uncertain
[4] J. B. Pearson, “Compensator design for dynamic optimization,” Int. J. systems,” Int. J. Control, vol. 66, no. 4, pp. 487–497, 1997.
Control, vol. 9, no. 4, pp. 473–482, 1969. [33] , “Robust pole assignment in uncertain systems,” Proc. Inst. Elec.
[5] F. M. Brasch and J. B. Pearson, “Pole placement using dynamic compen- Eng., vol. 144, no. 3, pp. 217–224, 1997.
sators,” IEEE Trans. Automat. Contr., vol. AC-15, pp. 34–43, Jan 1970.
[6] C.-T. Chen, “Design of pole placement compensators for multivariable
systems,” in Proc. Joint Automatic Control Conf., Atlanta, GA, 1970,
pp. 497–503.
[7] C.-T. Chen and C.-H. Hsu, “Design of dynamic compensators for mul-
tivariable systems,” in Proc. Joint Automatic Control Conf., 1971, pp.
893–900. Mehmet Turan Söylemez (S’98–A’99) was born
[8] H. Seraji, “An approach to dynamic compensator design for pole assign- in Darende, Turkey, in 1971. He received the B.Sc.
ment,” Int. J. Control, vol. 21, no. 6, pp. 955–966, 1975. degree in control and computer engineering from
[9] N. Munro and S. Novin-Hirbod, “Pole assignment using full-rank Istanbul Technical University (ITU), Turkey, the
output-feedback compensators,” Int. J. Syst. Sci., vol. 10, no. 3, pp. M.Sc. degree in control engineering and information
285–306, 1979. technology from the University of Manchester
[10] E. J. Davison and S.-H. Wang, “Properties of linear time-invariant mul- Institute of Science and Technology (UMIST), U.K.,
tivariable systems subject to arbitrary output and state feedback,” IEEE and the Ph.D. degree in control engineering from
Trans. Automat. Contr., vol. AC-18, pp. 24–32, Jan 1973. Control Systems Center, UMIST, in 1991, 1994, and
[11] R. Ahmari and A. G. Vacroux, “On the pole assignment in linear sys- 1999, respectively.
tems with fixed order compensators,” Int. J. Control, vol. 17, no. 2, pp. Since spring 2000, he has been with the Electrical
397–404, 1973. Engineering Department of Istanbul Technical University as an Assistant Pro-
[12] F. Fallside and H. Seraji, “Pole-shifting procedure for multivariable sys- fessor. He also gives lectures on signal processing and strategic website tech-
tems using output feedback,” Proc. lnst. Elec. Eng., vol. 118, no. 11, pp. nologies at Yeditepe University, Istanbul. His research interests include inverse
1648–1654, 1971. eigenvalue problems (pole assignment), multivariable systems, robust control,
[13] R. V. Patel, “Design of dynamic compensators for pole assignment,” Int. computer algebra, numerical analysis, genetic algorithms, PID controllers, and
J. Syst. Sci., vol. 7, no. 2, pp. 207–224, 1976. low-order controller design.
Authorized licensed use limited to: WASEDA UNIVERSITY LIBRARY. Downloaded on August 28,2023 at 05:26:03 UTC from IEEE Xplore. Restrictions apply.
SÖYLEMEZ AND MUNRO: A PARAMETRIC SOLUTION TO THE POLE ASSIGNMENT PROBLEM 723

Neil Munro received the B.Sc. degree in electrical

engineering from Strathclyde University, U.K.,
the M.Sc. and Ph.D. degrees from University of
Manchester Institute of Science and Technology
(UMIST), U.K., and the D.Sc. degree in control
engineering from the University of Manchester, U.K.
In 1979, he was appointed Professor of Applied
Control Engineering at UMIST, and has been Head
of the Control Systems Centre for 14 years. He spent
seven years in industry, working on submarine radar
systems, radar guidance systems for ground-to-air
missiles, industrial liquid metering controllers, and the design of alpha-numeric
displays. His current research interests are the development of CAD techniques
for the robust control of uncertain multivariable systems, using symbolic
algebra. Application areas include aerospace and process control.

Authorized licensed use limited to: WASEDA UNIVERSITY LIBRARY. Downloaded on August 28,2023 at 05:26:03 UTC from IEEE Xplore. Restrictions apply.